Indoor positioning with maximum likelihood classification of Wi-Fi signals

There are many applications for indoor positioning, from navigating hospitals, airports and shopping malls, to mining and disaster response. Despite intensive research, the problem of developing an accurate indoor location system remains largely unsolved. GPS is widely used for outdoor navigation, but its signals are attenuated and useless indoors. Augmented GPS, ultrasound, and inertial navigation systems have been proposed but remain impractical. The solution presented in this paper makes use of commonly available Wi-Fi networks. Implemented as an Android app on an ordinary smart phone, it comprises a calibration stage and a navigation stage. In the calibration stage, the system creates a Wi-Fi fingerprint for each room of a building, where the received signal power of multiple signals are collected over time and space and stored as multivariate Gaussian distributions. During the navigation stage, the system determines its position by matching Wi-Fi signal strengths to the fingerprints with maximum-likelihood classification. Evaluation results in home and commercial environments show that this classification method outperforms the more conventional nearest neighbor algorithm. The resulting system can determine its position in only a few seconds.